Danny Loeb

Computer Interpretation
of Diplomacy Openings

The Observation Module applied
to Spring 1995 in "dippouch"

As an example of the use of the observation module described in the article devoted to it elsewhere in this issue, and as a source of practical advice for potential "investors," let us apply the observation module to the results of the first turn of the game "dippouch", as shown in the first issue of this magazine.

General Comments about Spring 1901

In the previous issue of this magazine, I discussed how an opening book (such as the one published in this issue of Diplomatic Pouch) can be used to select a good opening.

Similarly, looking up your opponents openings in such a book gives you a good idea about kind of alliances they are likely to indicate. For example, in the game "dippouch," here is a brief summary of each country's opening. Click on the country's name for full details.

Austria. Balkan Gambit, Budapest Variation. Pro-Italy. Could be used against Russia or Turkey.
England. Wales Opening, Southern Opening. Anti-France.
France. Burgundy Opening. Basically defensive.
Germany. Blitzkreig Opening, Holland Variation. Defensive. Possibly anti-French.
Italy. Lepanto if convoy in Fall. Usually pro-Austria, anti-Turkey.
Russia. Turkish Attack. Usually just what it sounds like.
Turkey. Variant of Bosphorus opening (inferior, as it doesn't allow a build in Ank in case of Sev-Bla). Anti-Italy.

The openings library supplemented perhaps by a systematic search of games containing similar openings can occasionally give clues as to what kind of tell-tale moves to look for in Fall 1901 as well!

However, we hope that our diplomat's general operating procedures will be robust enough to work well in all phases including Spring 1901. Thus, we plan to use the normal observation routine even in Spring 1901 as described below. It is interesting however to use the openings library as a tool by which to judge the performance of the Observation Module.

One of the inputs of the Observation Module is the Friendliness Matrix produced in the previous season. Since this doesn't yet exist in Spring 1901, it must be invented. Below we take the all zero matrix as a default, however there are other possibilities.

Adjust the matrix in function of the pre-Spring 1901 negotation.
Use a predetermined matrix which compensates for "unjust." For example, moves by Germany and France towards the lowlands are considered by the observation module to be tantamount to attacks on each other. Similarly, Turkey, Austria, and Russia's moves to the Balkans are considered "aggressive."
Use a predetermined matrix to take take into account long term interests which the diplomat could otherwise not fathom. For example, Italy and Austria can often obtain a good short-term gain by stabbing each other. However, their long-term interest is often served by working together.

Conversion to DPP Protocol

Using a conversion program developed in LCS by Per Westling and myself, Diplomacy Adjudicator (judge) results can be directly converted to the DPP protocol language which is understood by the Bordeaux diplomat. First, I saved the results in a file called "dippouch". Then, I load the package in LCS

% lcs
LCS Project - CNRS/LAAS - 1992 
lcs version 3.1,  created 6-10-1992
val it = () : unit

- use "convert.pw.sml";
.....

convert.pw.sml file loaded
val it = () : unit

- val s = convert "dippouch";
val s = 
"ORD (SPR 101)
((AUS AMY BUD) MTO SER VLD) ((AUS AMY VIE) MTO BUD VLD) ((AUS FLT TRI) MTO ALB VLD)
((ENG AMY LVP) MTO WAL VLD) ((ENG FLT EDI) MTO NTH VLD) ((ENG FLT LON) MTO ENG VLD)
((FRA AMY MAR) MTO SPA VLD) ((FRA AMY PAR) MTO BUR VLD) ((FRA FLT BRE) MTO MAO VLD)
((GER AMY BER) MTO KIE VLD) ((GER AMY MUN) MTO RUH VLD) ((GER FLT KIE) MTO HOL VLD)
((ITA AMY ROM) MTO APU VLD) ((ITA AMY VEN) HLD VLD) ((ITA FLT NAP) MTO ION VLD)
((RUS AMY MOS) MTO SEV VLD) ((RUS AMY WAR) MTO UKR VLD) ((RUS FLT SEV) MTO BLA VLD) ((RUS FLT (STP SCS)) MTO BOT VLD)
((TUR AMY CON) MTO BUL VLD) ((TUR AMY SMY) MTO ANK VLD) ((TUR FLT ANK) MTO CON VLD)
EOM" : 
string

Observation Module

We then load the observation module implemented in LCS by Arnaud Moulard and Christophe Moustier.

- needs "load_all";
........
- observer nullMat77 (detail (ref2ord s));
val it = 
([[0.0, ~0.104166666666667, 0.9375, 3.9625, 0.0625, ~6.92083333333333,   ~1.57083333333333],
  [~0.104166666666667, 0.0, ~8.9, ~3.775, 0.345833333333333,   0.0541666666666667, ~0.0458333333333333],
  [0.9375, ~8.9, 0.0, ~5.70416666666667, ~6.82916666666667, ~0.8375,   ~0.879166666666667],
  [3.9625, ~3.775, ~5.70416666666667, 0.0, ~4.95416666666667,   ~1.25833333333333, 5.1875],
  [0.0625, 0.345833333333333, ~6.82916666666667, ~4.95416666666667, 0.0,   ~3.6, ~6.875],
  [~6.92083333333333, 0.0541666666666667, ~0.8375, ~1.25833333333333,   ~3.6, 0.0, ~11.8416666666667],
  [~1.57083333333333, ~0.0458333333333333, ~0.879166666666667, 5.1875,   ~6.875, ~11.8416666666667, 0.0]],
 [["AUS", "ITA"], ["ENG"], ["FRA"], ["GER", "RUS"], ["TUR"]]) : 
real list list * string list list

Or again

        A     E     F     G     I     T     R
    A        -0.1  +0.9  +4.0  +5.1  -5.9  -1.6 
    E  -0.1        -8.9  -3.8  +0.0   0.0  -0.0 
    F  +0.9  -8.9        -5.9  -0.0  +0.0  -1.1
    G  +4.0  -3.8  -5.9        +2.1  -0.4  +5.0
    I  +5.1  +0.0  -0.0  +2.1        -2.1  +0.1 
    T  -5.9   0.0  +0.0  -0.4  -2.1       -12.0 
    R  -1.6  -0.0  -1.1  +5.0  +0.1 -12.0

The results from only one turn of play do not yet show very strong relationships between the players. However, the computer considers England and France, and Russia and Turkey to be at war, and Austria and Italy, and Russia and Germany to be allies.

These results are based on the simple analysis of the results described below. In particular, none of the press or negotiations were taken into account.

How were these numbers generated

Many factors did not come into play...

Violation of Specific Agreements. No agreements were used in the above computation. If they had, then a positive entry would have been made for each treaty respected (such as that between France-Italy) and a negative entry for each treaty violated (such as that between France-England).
Coordinated Moves. No coordinated moves (supports or convoys) were made this season.
Attacks. There were no conflicts this season.
History. The all zero matrix was used to represent the history of the game, since this is the first season of the game. In practice, some other matrix might be used in order to give the diplomat a prejudice to consider that certain countries might be allied together.

The one important factor which came up in the calculations was the following.

Movement. All distances are computed with respect to current supply centers using the filter above.
- AUS BUD-SER moved towards Turkey -2
- AUS VIE-BUD moved away from France +1, Germany +2, and towards Turkey -1
- AUS TRI-ALB moved away from Italy +7, towards Turkey -1.
- ENG LVP-WAL is not counted as moving towards France, since the observation module does not take possible convoys into account.
- ENG EDI-NTH moved towards Germany -2 and France -2.
- ENG LON-ECH moves towards France -7.
- FRA MAR-SPA moves away from Austria 1, Germany 2, Italy 2.
- FRA PAR-BUR moves towards Austria -1, Germany -7, Italy -1, Russia -1
- FRA BRE-MAO moves towards Italy -1
- GER BER-KIE moves away from Russia 2
- GER MUN-RUH moves away from Austria 2, Italy 2, Russia 2
- GER KIE-HOL moves towards England -2, France -1
- ITA ROM-APU is distance neutral
- ITA NAP-ION moves towards Austria -2, Turkey -2
- RUS MOS-SEV moves towards Austria -2, Turkey -2, aways from Germany 1
- RUS WAR-UKR moves away from Germany 2, towards Turkey -1
- RUS SEV-BLA moves towards Turkey -7
- RUS (STP SCS)-BOT moves towards Germany -2
- TUR CON-BUL moves towards Austria -2, Italy -1, Russia -2
- TUR SMY-ANK is distance neutral
- TUR ANK-CON moves towards Italy -1
giving the following totals
```
        A     E     F     G     I     T     R
    A               1     2     7    -4 
    E              -9    -2 
    F                    -5                -1
    G   2    -2    -1           2           4
    I  -2                            -2
    T  -2                      -2          -2 
    R  -2                            -10
```

Symmetry This matrix is symmetrized giving

        A     E     F     G     I     T     R
    A               1     4     5    -6    -2
    E              -9    -4 
    F   1    -9          -6                -1
    G   4    -4    -6           2           5
    I   5                 2          -2
    T  -6                      -2         -12 
    R  -2          -1     5         -12

Transitivity. Finally, we added 0.004167 times the sqaure of the matrix to the original matrix.

Parameters

The following is a list of parameters used by the observation module and their default values used in this example. Actual optimal values can only be found through extensive testing (self-play).

Transitivity factor = 20.0
Old friendliness = 70%
Bounce = 5.0
Cut support = 10.0
Intent support = 6.0
Disrupt convoy = 10.0
Failed = 2.0
No move = 0.5
No such order = 3.0
Retreat = 20.0
Valid = 13.0
Distance Filter
- distance 0 = 30,
- distance 1 = 10,
- distance 2 = 3,
- distance 3 = 1,
- distance 4 or more = 0.
Limit Ally = 5.0
Limit Enemy = -7.0
Danny Loeb
Universite de Bordeaux I
(loeb@delanet.com)
If you wish to e-mail feedback on this article to the author, and clicking on the mail address above does not work for you, feel free to use the "Dear DP..." mail interface, which is located here....